stochastic_discrete_mlp_q_function.py 文件源码

def init_policy(self):
        output_vec = L.get_output(self._output_vec_layer, deterministic=True) / self._c
        prob = tf.nn.softmax(output_vec)
        max_qval = tf.reduce_logsumexp(output_vec, [1])

        self._f_prob = tensor_utils.compile_function([self._obs_layer.input_var], prob)
        self._f_max_qvals = tensor_utils.compile_function([self._obs_layer.input_var], max_qval)

        self._dist = Categorical(self._n)